Analyzing systemic lupus erythematosus publications using neural network-based multi-label classification algorithms.

Abstract

The heterogeneity in systemic lupus erythematosus research topics poses a challenge for the entire lupus community, from basic geneticists to clinical investigators. As such, it is critical for medical professionals to remain up to date on directions in lupus research and the main fields in which this research is being conducted (e.g., etiology, diagnosis, treatment, and outcomes). This article develops two multi-label text-classification models using Deep Neural Networks and Convolutional Neural Networks to classify the human-based adult-onset lupus–related articles in the PubMed database based on their abstract, keywords, and MeSH terms. During training evaluation, our models correctly indicated all relevant labels for 70% of the articles. The applied machine learning models (Deep Neural Network and Convolutional Neural Network) yielded a Micro-F1 score of 0.89, meaning that it successfully labeled the most relevant medical domains and types. In addition, these types of studies help the researchers be aware of the essential topics in this field, but due to difficulties in designing, the related studies are ignored or fade.